Foundry Advisory

We are looking for senior advisory input on Microsoft Foundry, starting with a short screening call as soon as possible. Context An enterprise environment with heavy bottom-up adoption of assistant tooling: a large number of individually-built agents and workflows sitting outside any managed environment. The organization wants to bring this under central governance without disrupting how people already work. Microsoft-centric estate. Fabric in place. Regulated sector, so audit, confidentiality and access control matter as much as capability. The question Our working position is that seat-based conversational tooling and consumption-based agent platforms should coexist, split by workload rather than by preference: Interactive, exploratory, human-in-the-loop work → seat-based licensing Scheduled or triggered automation → Foundry Anything needing audit, evaluation or versioning → Foundry The economics here are about pricing model rather than per-token rate. Seat licensing suits high-volume interactive use; consumption billing suits unattended automation. MCP looks like the bridge: connectors built for one platform should be consumable from the other, making this a migration into governance rather than a rebuild. We want that framing pressure-tested by someone who has done it. Topics for the call Production experience with Foundry specifically — agents, evaluation, observability. What broke that you did not expect? Where you draw the line between conversational tooling and agent platforms. Your read on the orchestration roadmap: Workflows retirement and the move to code-first. Build now or wait? MCP across platforms — what transfers cleanly and what does not. Per-agent cost attribution across multiple teams: separate projects and resource groups, or tags and telemetry? Reliability patterns for tool-grounded retrieval in multi-step agents. RBAC design at scale: environment owner, project manager, developer, consumer — and avoiding inherited role sprawl. Migration approach for a large inventory of existing agents: triage, prioritization, and bringing externally-built agents into a managed registry. Engagement Open shape. Initial advisory, with possible ongoing build support depending on fit. Happy to go into specifics under NDA.